As Expedia Group aims to operate as an AI-first company, its first chief AI and data officer, Xavier Amatriain, said the technology is redefining what’s possible for the industry’s products and services.
He joined the organization from Google late last year in the newly created role, with a heavy focus on the data foundation for the company’s technologists. Many large corporations have added a dedicated chief AI role to the C-suite, but more important than the role is dedicated ownership over AI, a report by Wharton Human-AI Research and research and analytics firm GBK Collective, found this week.
Expedia Group operates as a two-sided marketplace, working with both customers who are booking travel, and suppliers like hotels, airlines and rental stays.
Trust is essential in the travel industry, where customers are often planning big life moments, like a long-awaited family trip or honeymoon. It’s why the data foundations and careful experimentation are essential in the industry, Amatriain said.
“We believe that data is really a differentiator and a mode for us,” Amatriain said. “And it’s really important to keep it very connected to AI.”
Internal AI use
Travelers might be familiar with AI in Expedia’s customer-facing tools, but Expedia’s internal use-cases change often and are tested frequently, Amatriain said.
The chief AI and data officer said he follows a concept called “fanning out and fanning in,” trialing different AI tools, models and agents to see what might work and consolidating resources around the projects that show value.
“We’re very thoughtful about keeping at least 80%, if not more, of the people on what we call the golden path, but then allowing for others to experiment,” he said.
Those experimenters are called AI champions, employees advanced in their AI capabilities who often trial new tools and give feedback before they’re released to the wider staff. It’s easy for teams to get stuck once they’ve found a model that works for them, Amatriain said, but the quickly evolving technology constantly presents better, more efficient models.
“They become sort of like the people that actually bring the rest of the teams along, and we use them as educators, but also as evangelists of these new technologies,” Amatriain said.
Expedia Group also hosts education days on new technologies and tracks different metrics about usage and what works and doesn’t work for its champion employees. Amatriain warned against trying to track ROI as a static metric, because the technologies are so quickly evolving.
Amatriain said he works closely with the company’s chief information and security officer on governance, and works with the chief product officer to understand how AI is being built into Expedia’s features and to assess how it improves the product and the business.
The tech stack
When it comes to building a stack of AI models, Amatriain told CIO Dive the more the merrier.
“We use frontier, proprietary models, but we also use open weight models, and there’s a good use case for each of them,” he said.
The team is constantly evaluating tradeoffs between efficiency, intelligence and cost, the last of which is top of mind for chief tech officers who more often need to tie business value to AI outcomes.
Flexibility, optionality and avoiding vendor lock-in are of utmost importance in building their AI stack, Amatriain said.
“I’ve been saying this for a while, even when I was at Google, that I think models are going to be fungible, right?” Amatriain said. “You don’t need to get hooked into any specific model.”
Whatever model works well for a team today is “definitely going to be wrong next month,” he added, saying chief tech officers need to constantly anticipate change.
Strong infrastructure and culture is essential to adapt the tech stack and tech strategy quickly, Amatriain said. Tech teams need to be able to switch models behind the scenes, run constant evaluations and make sure the organization is achieving better results.
Reorganizing team structure
Expedia Group will lay off dozens of corporate employees at its Seattle headquarters through the end of this year, including data scientists and software development engineering roles. The cuts will also include multiple vice presidents, per an internal memo, which cited AI as part of the decision.
It is among a collection of other companies in the hospitality industry, including Mews and Marriott International, that have announced layoffs this year related to changes in roles because of AI.
The principles of flexibility for the company’s tech stack apply to its tech team as well, Amatriain said. The company aims to design organizations in an “anti-fragile way that not only are robust to change, but also benefit from change.”
The lines between tech roles are blurring, Amatriain said, as the responsibilities of engineers, product managers and scientists become closer while working with AI. It requires the company to move some pieces around, he said.
“We’re setting up the teams to be AI-first and to have this notion of flexibility and velocity,” Amatriain said. “This notion of a mission complete squad with very few dependencies across different organizations is something that is working very well.”
